Server has 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (list_, reply_, resolve_). However, tool descriptions are minimal (10-50 chars), lacking context on WHEN to use each tool and what they return. Parameters have types and descriptions, but lack constraints (enums, ranges). Output schemas are not documented, LLMs cannot predict response structure. The 'reply_and_resolve' tool violates single-responsibility principle by combining two actions. Error handling is generic ('Error: {str(e)}') with no recovery guidance.

• Expand tool descriptions to 100-200 chars. Include: what the tool does, when to use it (vs similar tools), prerequisites, and what it returns. Example: 'List unresolved review threads on a GitHub PR. Call this first to see pending feedback before replying. Returns thread IDs, file paths, and comment counts.'
• Split 'reply_and_resolve' into two separate tools: 'reply_to_review_thread' (already exists) and 'resolve_review_thread' (already exists). Remove the combined tool to avoid agent confusion.
• Document output schemas in tool descriptions or as a separate 'outputSchema' field. Specify: field names, types, and whether lists are paginated. Example: 'Returns {threads: [{id, path, line, isResolved, commentCount}], totalCount}'.
• Replace generic error handling with categorized, actionable messages. Example: 'Thread not found (ID: xyz). Verify the PR number and call list_review_threads() to get valid thread IDs.' Distinguish retryable vs fatal errors.
• Add parameter constraints: 'body' should have maxLength (e.g. 65536 chars); 'pull_number' should have minimum 1. Document in descriptions: 'Pull request number (must be > 0)'.